This project uses AdaBoost classifiers to detect pedestrians in still images. A two-stage approach is adopted for acceleration - a simple first-stage AdaBoost classifier is used to eliminate most negative samples and then a more complex second-stage one is used to detect pedestrians. The weak learners for AdaBoost classifiers are two-level decision trees whose nodes are labeled with integral channel features computed from LUV channels, gradient magnitude, and gradient magnitude with six orientations. The numbers of weak learners for the first and second stage classifiers are 100 and 2,000 respectively.

for research use only

1. OpenCV 2.4.9
2. build-essential

Ubuntu 14.04 LTS

1. first-stage: model/boost_weak100_pos23266_neg15624_icf780_obj80x160_cell12_12_4_1.xml
2. second-stage: model/boost_weak2000_pos23266_neg15624_icf9240_obj80x160_cell12_16_4_2.xml;

1. navigate into project root directory
2. run the command 'make' to build libpedetect.a (located in project_dir/lib/)

1. navigate into sample/detection/
2. run 'build.sh'
3. run './detect_image' (this program detects pedestrians in sample/detection/sample.jpg and saves the result in sample/detection/result.jpg)